Booking Holdings is spending roughly $8 billion to $9 billion a year on paid customer-acquisition channels as it tries to stay in front of travelers in a market being reshaped by artificial intelligence.
Ewout Steenbergen, EVP and CFO of Booking Holdings, said at the Fortune AIQ Summit that the company is using AI to make travel planning more seamless, personalized and proactive. He described scenarios in which AI could suggest an indoor visit to the Louvre if rain threatens a Paris itinerary, move a restaurant reservation after a delayed flight, or recommend an activity based on a traveler’s preferences and open time.
The company’s goal is not just a better chatbot. Booking is pushing toward what it calls a 'connected trip' — an integrated experience that brings together flights, hotels, restaurants, rental cars and activities. Steenbergen said that could increase loyalty, trust and frequency of engagement.
For finance leaders, the question is whether AI improves the economics, not whether it looks impressive in a demo. Steenbergen said Booking is watching unit economics closely and does not have a perfect answer for return on investment. He said even the large hyperscalers spending hundreds of billions on large language models do not really know the ROI.
Early signs inside Booking are positive. Steenbergen said customers using its AI tools are booking somewhat faster, converting at somewhat higher rates and canceling less often, though he cautioned that the data is still early stage.
Booking’s traffic mix shows why channel economics matter. Steenbergen said about one-third of consumer traffic comes through paid channels, while the other two-thirds comes directly to its platforms. As consumers turn to more AI tools and agents for discovery, he said greater channel diversification could help Booking’s unit economics.
The company is also trying to rebuild workflows end to end rather than simply layer AI onto old processes. In customer service, Steenbergen said booking units have grown at a high single-digit rate while total customer-service costs declined slightly, with customer satisfaction rising at the same time. That pushed down average servicing cost per booking.
In engineering, Booking’s roughly 9,000 technologists have used AI tools to increase the number of merge requests put into production by about 30%, according to Steenbergen. He said that figure reflects code that has been tested and cleared quality controls, not just generated.
The CFO said the company is also measuring the cost side carefully. Booking monitors token and licensing costs, uses 'effective model-cost routing' and reserves lower-cost or open-source models for simpler tasks while using more expensive models for complex ones. One engineering metric combines human and AI costs: total IT cost per merge request.
Steenbergen is using the tools himself as well. He said he has an 'AI coach' and uses two AI agents: a 'strategic thought partner' for board materials and strategic analysis, and a 'critical equity research agent' to pressure-test earnings-call preparation.
The numbers come first. Booking is spending billions to win attention, while trying to prove that AI can lower the all-in cost of serving each traveler. Capital rewards clear rules, and this is the kind of test the market understands: better conversion, lower cancellations and tighter margins, or another expensive technology cycle that burns cash before it earns trust.
